Abstract
In this paper, we propose a cylindrical stitching method which based on template frame image to realize 360° panoramic video stitching. Firstly, we selected template frame to determine the camera parameters and distortion coefficient. And, removing the image distortion. After that, extracting the template frame feature and making rough matching with the feature points, and then eliminating the feature points of mismatching by the improved feature point screening method. Then, feature points are projected on the cylinder in advance for subsequent video image fusion. Finally, image fusion is carried out after the cylinder projection of the video image. The speed was greatly improved in the subsequent video splicing by using camera parameters, distortion coefficient and transformation matrix of template frame image. In addition, the CUDA platform was adopted for GPU parallel processing in order to further improve the speed to meet the real-time requirements. Finally, we realize real-time video stitching with 30 frames per second.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
